{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:6LJO423GK7BICIA37XSAA6QSCA","short_pith_number":"pith:6LJO423G","schema_version":"1.0","canonical_sha256":"f2d2ee6b6657c281201bfde4007a12101bf8aaca4bbdf0ca3ec7e8f66c6345b3","source":{"kind":"arxiv","id":"2505.17548","version":1},"attestation_state":"computed","paper":{"title":"H2:Towards Efficient Large-Scale LLM Training on Hyper-Heterogeneous Cluster over 1,000 Chips","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.DC","authors_text":"Ding Tang, Huihuang Zheng, Hui Wang, Jiakai Hu, Jiecheng Zhou, Shengwei Li, Xingcheng Zhang, Zhilin Pei","submitted_at":"2025-05-23T06:54:29Z","abstract_excerpt":"Recent advancements in large language models (LLMs) necessitate extensive computational resources, prompting the use of diverse hardware accelerators from multiple vendors. However, traditional distributed training frameworks struggle to efficiently utilize hyper-heterogeneous clusters comprising thousands of chips due to significant disparities in software stacks, operator implementations, communication libraries, and hardware capabilities. To address these challenges, we propose H2, which stands for HyperHetero and is a systematic framework enabling efficient training of LLMs on clusters wit"},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2505.17548","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by-sa/4.0/","primary_cat":"cs.DC","submitted_at":"2025-05-23T06:54:29Z","cross_cats_sorted":[],"title_canon_sha256":"220821ff60465e455efc48f9f88a087e55e2c6c624bd3fa9ff57ab02413923f2","abstract_canon_sha256":"e56408a9e0fa9aa0b0de396917ea3b3dc7113f3e906a4696660a9cd5528a5a46"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:08:28.033781Z","signature_b64":"+CINiBdxWClIOE9MxDj82vEbeoR+FEPaoUkb99H9haeFkjDx1247SmtCZeYYJerkZp+Jwh7/Iqss/TctQcR1Cg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"f2d2ee6b6657c281201bfde4007a12101bf8aaca4bbdf0ca3ec7e8f66c6345b3","last_reissued_at":"2026-07-05T11:08:28.033301Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:08:28.033301Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"H2:Towards Efficient Large-Scale LLM Training on Hyper-Heterogeneous Cluster over 1,000 Chips","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.DC","authors_text":"Ding Tang, Huihuang Zheng, Hui Wang, Jiakai Hu, Jiecheng Zhou, Shengwei Li, Xingcheng Zhang, Zhilin Pei","submitted_at":"2025-05-23T06:54:29Z","abstract_excerpt":"Recent advancements in large language models (LLMs) necessitate extensive computational resources, prompting the use of diverse hardware accelerators from multiple vendors. However, traditional distributed training frameworks struggle to efficiently utilize hyper-heterogeneous clusters comprising thousands of chips due to significant disparities in software stacks, operator implementations, communication libraries, and hardware capabilities. To address these challenges, we propose H2, which stands for HyperHetero and is a systematic framework enabling efficient training of LLMs on clusters wit"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.17548","kind":"arxiv","version":1},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2505.17548/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"aliases":[{"alias_kind":"arxiv","alias_value":"2505.17548","created_at":"2026-07-05T11:08:28.033368+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.17548v1","created_at":"2026-07-05T11:08:28.033368+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.17548","created_at":"2026-07-05T11:08:28.033368+00:00"},{"alias_kind":"pith_short_12","alias_value":"6LJO423GK7BI","created_at":"2026-07-05T11:08:28.033368+00:00"},{"alias_kind":"pith_short_16","alias_value":"6LJO423GK7BICIA3","created_at":"2026-07-05T11:08:28.033368+00:00"},{"alias_kind":"pith_short_8","alias_value":"6LJO423G","created_at":"2026-07-05T11:08:28.033368+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.07569","citing_title":"HexiSeq: Accommodating Long Context Training of LLMs over Heterogeneous Hardware","ref_index":51,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/6LJO423GK7BICIA37XSAA6QSCA","json":"https://pith.science/pith/6LJO423GK7BICIA37XSAA6QSCA.json","graph_json":"https://pith.science/api/pith-number/6LJO423GK7BICIA37XSAA6QSCA/graph.json","events_json":"https://pith.science/api/pith-number/6LJO423GK7BICIA37XSAA6QSCA/events.json","paper":"https://pith.science/paper/6LJO423G"},"agent_actions":{"view_html":"https://pith.science/pith/6LJO423GK7BICIA37XSAA6QSCA","download_json":"https://pith.science/pith/6LJO423GK7BICIA37XSAA6QSCA.json","view_paper":"https://pith.science/paper/6LJO423G","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.17548&json=true","fetch_graph":"https://pith.science/api/pith-number/6LJO423GK7BICIA37XSAA6QSCA/graph.json","fetch_events":"https://pith.science/api/pith-number/6LJO423GK7BICIA37XSAA6QSCA/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/6LJO423GK7BICIA37XSAA6QSCA/action/timestamp_anchor","attest_storage":"https://pith.science/pith/6LJO423GK7BICIA37XSAA6QSCA/action/storage_attestation","attest_author":"https://pith.science/pith/6LJO423GK7BICIA37XSAA6QSCA/action/author_attestation","sign_citation":"https://pith.science/pith/6LJO423GK7BICIA37XSAA6QSCA/action/citation_signature","submit_replication":"https://pith.science/pith/6LJO423GK7BICIA37XSAA6QSCA/action/replication_record"}},"created_at":"2026-07-05T11:08:28.033368+00:00","updated_at":"2026-07-05T11:08:28.033368+00:00"}